In pattern transfer, as in any other method of information transfer, the output is usually a nonlinear function of the input. Lithography at the limit of resolution is an excellent object to demonstrate this. Printing structures smaller than 300 nm with a 4 X 0.5NA tool, the derivative of the pattern transfer function, or the ratio of pattern size variations on the wafer over pattern size variations at the mask level, is not a 4:1, as one would expect from the demagnification of the step and scan tool. In other words, below 300 nm, mask linewidth variations (for example butting errors of the mask writing tool) print at about twice their expected size. In the concept of the pattern transfer function, a mask defect is viewed as a localized variation in the linewidth of the mask. The printing of a mask defect therefore depends strongly on the slope of the pattern transfer function. Defects smaller than 200 nm on the mask already cause a significant linewidth variation on the wafer, if those defects are in a regular array of 250 nm lines/300 nm spaces or in 300 nm contact holes. Lithogrpahy in a manufacturing environment means to deliver the designed pattern over large areas using real masks. We discuss our strategies of how we try to minimize the influence of mask irregularities in 0.25 micrometers lithography for the development of the 256M DRAM. Although certain improvements are possible, the nonlinearity of the pattern transfer function at low k obviously demands extremely tight mask specifications beyond the limits of current tools and processes.